Hi, I'm trying to follow the docs at http://www.gentoo.org/doc/en/hpc-howto.xml. I've edited the file /usr/share/mpich/machines.LINUX but mpich expects that to be somewhere else! # vim /usr/share/mpich/machines.LINUX # tstmachines -v LINUX Cannot read list of nodes /usr/share/machines.LINUX # Regarding the documentation, do I really need /etc/exports file? I don't use NFS. Do I have to then? The docs should say that clearly. Also, is mpich an alternative to openpbs? So should I just follow one section of the howto file?
(In reply to comment #0) > Hi, > I'm trying to follow the docs at http://www.gentoo.org/doc/en/hpc-howto.xml. > I've edited the file /usr/share/mpich/machines.LINUX but mpich expects that to > be somewhere else! Which revision of mpich are you using? Please try 1.2.6-r2. > Regarding the documentation, do I really need /etc/exports file? I don't use > NFS. Do I have to then? The docs should say that clearly. No, you don't. And the doc isn't designed to teach you everything about all the related technologies to clustering, it's designed to get a cluster set up. If you want a full background on everything, you'll have to read the individual docs for each package. > Also, is mpich an > alternative to openpbs? So should I just follow one section of the howto file? No, mpich is not an alternative to a job batching system. But a batching system isn't required.
I use that version - sys-cluster/mpich-1.2.6-r2, see $Subj.
(In reply to comment #2) > I use that version - sys-cluster/mpich-1.2.6-r2, see $Subj. I'm using sys-cluster/mpich-1.2.6-r3 on AMD and the bug is still there. A symlink from /usr/share/mpich/machines.LINUX to /usr/share/machines.LINUX seems to work though.
Seems that it's ignoring the --datadir setting. I'm passing --datadir=/usr/share/mpich in configure, and e.g. mpirun shows up with datadir=/usr/share. Looking into a fix.
Should be fixed in -r4.
I haven't checked but maybe the fix should also be applied to 1.2.7_p1?
You're exactly right, you haven't checked. =P Please don't reopen bugs unless the problem actually isn't fixed. A diff, or a read of the ChangeLog that shows 1.2.7 being added after 1.2.6-r4, should suggest this is fixed in both.